2 * Author......: Jens Steube <jens.steube@gmail.com>
9 //#define NEW_SIMD_CODE
11 #include "inc_vendor.cl"
12 #include "inc_hash_constants.h"
13 #include "inc_hash_functions.cl"
14 #include "inc_types.cl"
15 #include "inc_common.cl"
18 #include "inc_simd.cl"
20 #define SIPROUND(v0,v1,v2,v3) \
22 (v1) = rotl64 ((v1), 13); \
24 (v0) = rotl64 ((v0), 32); \
26 (v3) = rotl64 ((v3), 16); \
29 (v3) = rotl64 ((v3), 21); \
32 (v1) = rotl64 ((v1), 17); \
34 (v2) = rotl64 ((v2), 32)
36 __kernel void m10100_m04 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
42 const u32 lid = get_local_id (0);
48 const u32 gid = get_global_id (0);
50 if (gid >= gid_max) return;
55 pw_buf0[0] = pws[gid].i[ 0];
56 pw_buf0[1] = pws[gid].i[ 1];
57 pw_buf0[2] = pws[gid].i[ 2];
58 pw_buf0[3] = pws[gid].i[ 3];
59 pw_buf1[0] = pws[gid].i[ 4];
60 pw_buf1[1] = pws[gid].i[ 5];
61 pw_buf1[2] = pws[gid].i[ 6];
62 pw_buf1[3] = pws[gid].i[ 7];
64 const u32 pw_len = pws[gid].pw_len;
70 u64x v0p = SIPHASHM_0;
71 u64x v1p = SIPHASHM_1;
72 u64x v2p = SIPHASHM_2;
73 u64x v3p = SIPHASHM_3;
75 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
76 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
77 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
78 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
84 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
91 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
99 case 0: w0[1] |= out_len << 24; break;
100 case 1: w0[3] |= out_len << 24; break;
101 case 2: w1[1] |= out_len << 24; break;
102 case 3: w1[3] |= out_len << 24; break;
113 for (i = 0, j = 0; i <= out_len && i < 16; i += 8, j += 2)
115 u64x m = hl32_to_64 (w0[j + 1], w0[j + 0]);
119 SIPROUND (v0, v1, v2, v3);
120 SIPROUND (v0, v1, v2, v3);
125 for ( j = 0; i <= out_len && i < 32; i += 8, j += 2)
127 u64x m = hl32_to_64 (w1[j + 1], w1[j + 0]);
131 SIPROUND (v0, v1, v2, v3);
132 SIPROUND (v0, v1, v2, v3);
139 SIPROUND (v0, v1, v2, v3);
140 SIPROUND (v0, v1, v2, v3);
141 SIPROUND (v0, v1, v2, v3);
142 SIPROUND (v0, v1, v2, v3);
144 const u64x v = v0 ^ v1 ^ v2 ^ v3;
146 const u32x a = l32_from_64 (v);
147 const u32x b = h32_from_64 (v);
151 COMPARE_M_SIMD (a, b, z, z);
155 __kernel void m10100_m08 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
159 __kernel void m10100_m16 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
163 __kernel void m10100_s04 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
169 const u32 lid = get_local_id (0);
175 const u32 gid = get_global_id (0);
177 if (gid >= gid_max) return;
182 pw_buf0[0] = pws[gid].i[ 0];
183 pw_buf0[1] = pws[gid].i[ 1];
184 pw_buf0[2] = pws[gid].i[ 2];
185 pw_buf0[3] = pws[gid].i[ 3];
186 pw_buf1[0] = pws[gid].i[ 4];
187 pw_buf1[1] = pws[gid].i[ 5];
188 pw_buf1[2] = pws[gid].i[ 6];
189 pw_buf1[3] = pws[gid].i[ 7];
191 const u32 pw_len = pws[gid].pw_len;
197 u64x v0p = SIPHASHM_0;
198 u64x v1p = SIPHASHM_1;
199 u64x v2p = SIPHASHM_2;
200 u64x v3p = SIPHASHM_3;
202 v0p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
203 v1p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
204 v2p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[1], salt_bufs[salt_pos].salt_buf[0]);
205 v3p ^= hl32_to_64 (salt_bufs[salt_pos].salt_buf[3], salt_bufs[salt_pos].salt_buf[2]);
211 const u32 search[4] =
213 digests_buf[digests_offset].digest_buf[DGST_R0],
214 digests_buf[digests_offset].digest_buf[DGST_R1],
223 for (u32 il_pos = 0; il_pos < il_cnt; il_pos += VECT_SIZE)
230 const u32x out_len = apply_rules_vect (pw_buf0, pw_buf1, pw_len, rules_buf, il_pos, w0, w1);
238 case 0: w0[1] |= out_len << 24; break;
239 case 1: w0[3] |= out_len << 24; break;
240 case 2: w1[1] |= out_len << 24; break;
241 case 3: w1[3] |= out_len << 24; break;
252 for (i = 0, j = 0; i <= out_len && i < 16; i += 8, j += 2)
254 u64x m = hl32_to_64 (w0[j + 1], w0[j + 0]);
258 SIPROUND (v0, v1, v2, v3);
259 SIPROUND (v0, v1, v2, v3);
264 for ( j = 0; i <= out_len && i < 32; i += 8, j += 2)
266 u64x m = hl32_to_64 (w1[j + 1], w1[j + 0]);
270 SIPROUND (v0, v1, v2, v3);
271 SIPROUND (v0, v1, v2, v3);
278 SIPROUND (v0, v1, v2, v3);
279 SIPROUND (v0, v1, v2, v3);
280 SIPROUND (v0, v1, v2, v3);
281 SIPROUND (v0, v1, v2, v3);
283 const u64x v = v0 ^ v1 ^ v2 ^ v3;
285 const u32x a = l32_from_64 (v);
286 const u32x b = h32_from_64 (v);
290 COMPARE_S_SIMD (a, b, z, z);
294 __kernel void m10100_s08 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)
298 __kernel void m10100_s16 (__global pw_t *pws, __global kernel_rule_t * rules_buf, __global comb_t *combs_buf, __global bf_t *bfs_buf, __global void *tmps, __global void *hooks, __global u32 *bitmaps_buf_s1_a, __global u32 *bitmaps_buf_s1_b, __global u32 *bitmaps_buf_s1_c, __global u32 *bitmaps_buf_s1_d, __global u32 *bitmaps_buf_s2_a, __global u32 *bitmaps_buf_s2_b, __global u32 *bitmaps_buf_s2_c, __global u32 *bitmaps_buf_s2_d, __global plain_t *plains_buf, __global digest_t *digests_buf, __global u32 *hashes_shown, __global salt_t *salt_bufs, __global void *esalt_bufs, __global u32 *d_return_buf, __global u32 *d_scryptV0_buf, __global u32 *d_scryptV1_buf, __global u32 *d_scryptV2_buf, __global u32 *d_scryptV3_buf, const u32 bitmap_mask, const u32 bitmap_shift1, const u32 bitmap_shift2, const u32 salt_pos, const u32 loop_pos, const u32 loop_cnt, const u32 il_cnt, const u32 digests_cnt, const u32 digests_offset, const u32 combs_mode, const u32 gid_max)